自然语言处理NLP相关文章合集,都由本人看过。欢迎来信探讨
LAMP-TR-135
CS-TR-4831
UMIACS-TR-2006-47
ASurveyofStatisticalMachineTranslation
AdamLopezApril2007
ComputationalLinguisticsandInformationProcessingLaboratory
InstituteforAdvancedComputerStudies
DepartmentofComputerScience
UniversityofMaryland
CollegePark,MD20742
alopez@cs.umd.edu
Abstract
Statisticalmachinetranslation(SMT)treatsthetranslationofnaturallanguageasamachinelearningproblem.Byexaminingmanysamplesofhuman-producedtranslation,SMTalgorithmsautomaticallylearnhowtotranslate.SMThasmadetremendousstridesinlessthantwodecades,andmanypopulartechniqueshaveonlyemergedwithinthelastfewyears.Thissurveypresentsatutorialoverviewofstate-of-the-artSMTatthebeginningof2007.Webeginwiththecontextofthecurrentresearch,andthenmovetoaformalproblemdescriptionandanoverviewofthefourmainsubproblems:translationalequivalencemodeling,mathematicalmodeling,parameterestimation,anddecoding.Alongtheway,wepresentataxonomyofsomedifferentapproacheswithintheseareas.Weconcludewithanoverviewofevaluationandnotesonfuturedirections.
Thisisareviseddraftofapapercurrentlyunderreview.Thecontentsmaychangeinlaterdrafts.Pleasesendanycomments,questions,orcorrectionstoalopez@cs.umd.edu.FeelfreetociteasUniversityofMarylandtechnicalreportUMIACS-TR-2006-47.ThesupportofthisresearchbytheGALEprogramoftheDefenseAdvancedResearchProjectsAgency,ContractNo.HR0011-06-2-0001,ONRMURIContractFCPO.810548265,andDepartmentofDefensecontractRD-02-5700isacknowledged.